Ayse S. Cakmak

Background: Heart failure (HF) is a major cause of frequent hospitalization and death. Early detection of HF symptoms using smartphone-based monitoring may reduce adverse events in a low-cost, scalable way. Objective: We examined the relationship of HF decompensation events with smartphone-based features derived from passively and actively acquired data. Methods: This was a prospective cohort study in which we monitored HF participants’ social and movement activities using a smartphone app and followed them for clinical events via phone and chart review and classified the encounters as compensated or decompensated by reviewing the provider notes in detail. We extracted motion, location, and social interaction passive features and self-reported quality of life weekly (active) with the short Kansas City Cardiomyopathy Questionnaire (KCCQ-12) survey. We developed and validated an algorithm for classifying decompensated versus compensated clinical encounters (hospitalizations or clinic visits). We evaluated models based on single modality as well as early and late fusion approaches combining patient-reported outcomes and passive smartphone data. We used Shapley additive explanation values to quantify the contribution and impact of each feature to the model. Results: We evaluated 28 participants with a mean age of 67 years (SD 8), among whom 11% (3/28) were female and 46% (13/28) were Black. We identified 62 compensated and 48 decompensated clinical events from 24 and 22 participants, respectively. The highest area under the precision-recall curve (AUCPr) for classifying decompensation was with a late fusion approach combining KCCQ-12, motion, and social contact features using leave-one-subject-out cross-validation for a 2-day prediction window. It had an AUCPr of 0.80, with an area under the receiver operator curve (AUC) of 0.83, a positive predictive value (PPV) of 0.73, a sensitivity of 0.77, and a specificity of 0.88 for a 2-day prediction window. Similarly, the 4-day window model had an AUC of 0.82, an AUCPr of 0.69, a PPV of 0.62, a sensitivity of 0.68, and a specificity of 0.87. Passive social data provided some of the most informative features, with fewer calls of longer duration associating with a higher probability of future HF decompensation. Conclusions: Smartphone-based data that includes both passive monitoring and actively collected surveys may provide important behavioral and functional health information on HF status in advance of clinical visits. This proof-of-concept study, although small, offers important insight into the social and behavioral determinants of health and the feasibility of using smartphone-based monitoring in this population. Our strong results are comparable to those of more active and expensive monitoring approaches, and underscore the need for larger studies to understand the clinical significance of this monitoring method.

Post-Traumatic Stress Disorder (PTSD) is a psychiatric condition resulting from threatening or horrifying events. We hypothesized that circadian rhythm changes, measured by a wrist-worn research watch are predictive of post-trauma outcomes. Approach: 1618 post-trauma patients were enrolled after admission to emergency departments (ED). Three standardized questionnaires were administered at week eight to measure post-trauma outcomes related to PTSD, sleep disturbance, and pain interference with daily life. Pulse activity and movement data were captured from a research watch for eight weeks. Standard and novel movement and cardiovascular metrics that reflect circadian rhythms were derived using this data. These features were used to train different classifiers to predict the three outcomes derived from week-eight surveys. Clinical surveys administered at ED were also used as features in the baseline models. Results: The highest cross-validated performance of research watch-based features was achieved for classifying participants with pain interference by a logistic regression model, with an area under the receiver operating characteristic curve (AUC) of 0.70. The ED survey-based model achieved an AUC of 0.77, and the fusion of research watch and ED survey metrics improved the AUC to 0.79. Significance: This work represents the first attempt to predict and classify post-trauma symptoms from passive wearable data using machine learning approaches that leverage the circadian desynchrony in a potential PTSD population.

Objective. To develop a sleep staging method from wrist-worn accelerometry and the photoplethysmogram (PPG) by leveraging transfer learning from a large electrocardiogram (ECG) database. Approach. In previous work, we developed a deep convolutional neural network for sleep staging from ECG using the cross-spectrogram of ECG-derived respiration and instantaneous beat intervals, heart rate variability metrics, spectral characteristics, and signal quality measures derived from 5793 subjects in Sleep Heart Health Study (SHHS). We updated the weights of this model by transfer learning using PPG data derived from the Empatica E4 wristwatch worn by 105 subjects in the ‘Emory Twin Study Follow-up’ (ETSF) database, for whom overnight polysomnographic (PSG) scoring was available. The relative performance of PPG, and actigraphy (Act), plus combinations of these two signals, with and without transfer learning was assessed. Main results. The performance of our model with transfer learning showed higher accuracy (1–9 percentage points) and Cohen’s Kappa (0.01–0.13) than those without transfer learning for every classification category. Statistically significant, though relatively small, incremental differences in accuracy occurred for every classification category as tested with the McNemar test. The out-of-sample classification performance using features from PPG and actigraphy for four-class classification was Accuracy (Acc) = 68.62% and Kappa = 0.44. For two-class classification, the performance was Acc = 81.49% and Kappa = 0.58. Significance. We proposed a combined PPG and actigraphy-based sleep stage classification approach using transfer learning from a large ECG sleep database. Results demonstrate that the transfer learning approach improves estimates of sleep state. The use of automated beat detectors and quality metrics means human over-reading is not required, and the approach can be scaled for large cross-sectional or longitudinal studies using wrist-worn devices for sleep staging.

Study Objectives: The usage of wrist-worn wearables to detect sleep–wake states remains a formidable challenge, particularly among individuals with disordered sleep. We developed a novel and unbiased data-driven method for the detection of sleep–wake and compared its performance with the well-established Oakley algorithm (OA) relative to polysomnography (PSG) in elderly men with disordered sleep. Methods: Overnight in-lab PSG from 102 participants was compared with accelerometry and photoplethysmography simultaneously collected with a wearable device (Empatica E4). A binary segmentation algorithm was used to detect change points in these signals. A model that estimates sleep or wake states given the changes in these signals was established (change point decoder, CPD). The CPD’s performance was compared with the performance of the OA in relation to PSG. Results: On the testing set, OA provided sleep accuracy of 0.85, wake accuracy of 0.54, AUC of 0.67, and Kappa of 0.39. Comparable values for CPD were 0.70, 0.74, 0.78, and 0.40. The CPD method had sleep onset latency error of −22.9 min, sleep efficiency error of 2.09%, and underestimated the number of sleep–wake transitions with an error of 64.4. The OA method’s performance was 28.6 min, −0.03%, and −17.2, respectively. Conclusions: The CPD aggregates information from both cardiac and motion signals for state determination as well as the cross-dimensional influences from these domains. Therefore, CPD classification achieved balanced performance and higher AUC, despite underestimating sleep–wake transitions. The CPD could be used as an alternate framework to investigate sleep–wake dynamics within the conventional time frame of 30-s epochs.